Re:A Skeptical Take on ChatGPT: Ezra Klein Interviews Gary M...
by mckravchyk on Saturday January 07, 2023 @04:34AM
(#34286614)
I have noticed that it makes up new API methods that don't exist on the fly if it fits the context. Another example, I mistakenly thought that a certain thing can be represented as a class instance, and I asked it to tell me how to get the class instance given the item id, and it came up with an API to do that, and then I asked further about operations on the instance and it came up with method names on a class that turned out to not exist. It's creative for sure.
[ Parent
| on HN ]
Re:A Skeptical Take on ChatGPT: Ezra Klein Interviews Gary M...
by arkitaip on Saturday January 07, 2023 @04:38AM
(#34286621)
Isn't it interesting how OpenAI has neutered ChatGPT to be insanely politically correct on even non-controversial and mundane topics yet they won't put the proper mechanisms in place so that ChatGPT can at least be correct about its most basic claims.
Re:A Skeptical Take on ChatGPT: Ezra Klein Interviews Gary M...
by notahacker on Saturday January 07, 2023 @08:06AM
(#34287671)
This assumes, of course, that getting an LLM optimised for chat to flawlessly parse and interpret novel computer programs is no more difficult than getting it to default to "I am sorry, but I cannot offer advice on..." style of boilerplated non-answer, slap caveats about asking trusted sources and good points not outweighing bad points, avoid naughty words and topics and prioritise answers which are similar to the mainstream stuff rather than the fringe stuff in its corpus...
Re:A Skeptical Take on ChatGPT: Ezra Klein Interviews Gary M...
by jnsaff2 on Saturday January 07, 2023 @08:02AM
(#34287649)
Yes! I saw this as well. I asked it to write boto3 to download all files in a s3 prefix and it happily made up a method that does exactly that, great, except it does not exist in boto3, you have to list objects with that prefix and iterate over it.